viztracer 0.17.1

A debugging and profiling tool that can trace and visualize python code execution

VizTracer

VizTracer is a low-overhead logging/debugging/profiling tool that can trace and visualize your python code execution.

The front-end UI is powered by Perfetto. Use "AWSD" to zoom/navigate. More help can be found in "Support - Controls".

Highlights

• Detailed function entry/exit information on timeline with source code
• Super easy to use, no source code change for most features, no package dependency
• Supports threading, multiprocessing, subprocess, async and PyTorch
• Powerful front-end, able to render GB-level trace smoothly
• Works on Linux/MacOS/Windows

Install

The preferred way to install VizTracer is via pip

pip install viztracer

Basic Usage

Command Line

# Instead of "python3 my_script.py arg1 arg2"
viztracer my_script.py arg1 arg2

A result.json file will be generated, which you can open with vizviewer

# You can display all the files in a directory and open them in browser too
vizviewer ./
# For very large trace files, try external trace processor
vizviewer --use_external_processor result.json

vizviewer will host an HTTP server on http://localhost:9001. You can also open your browser and use that address.

If you do not want vizviewer to open the webbrowser automatically, you can use

vizviewer --server_only result.json

If you just need to bring up the trace report once, and do not want the persistent server, use

vizviewer --once result.json

vizviewer result.json

A VS Code Extension is available to make your life even easier.

Add --open to open the reports right after tracing

viztracer --open my_script.py arg1 arg2
viztracer -o result.html --open my_script.py arg1 arg2

modules and console scripts(like flask) are supported as well

viztracer -m your_module

viztracer flask run

Inline

You can also manually start/stop VizTracer in your script as well.

from viztracer import VizTracer

tracer = VizTracer()
tracer.start()
# Something happens here
tracer.stop()
tracer.save() # also takes output_file as an optional argument

Or, you can do it with with statement

with VizTracer(output_file="optional.json") as tracer:
    # Something happens here

Jupyter

If you are using Jupyter, you can use viztracer cell magics.

# You need to load the extension first
%load_ext viztracer

%%viztracer
# Your code after

A VizTracer Report button will appear after the cell and you can click it to view the results

PyTorch

VizTracer can log native calls and GPU events of PyTorch (based on torch.profiler) with --log_torch.

with VizTracer(log_torch=True) as tracer:
    # Your torch code

viztracer --log_torch your_model.py

Advanced Usage

Trace Filter

VizTracer can filter out the data you don't want to reduce overhead and keep info of a longer time period before you dump the log.

• Min Duration
• Max Stack Depth
• Include Files
• Exclude Files
• Ignore C Function
• Sparse Log

Extra Logs without Code Change

VizTracer can log extra information without changing your source code

• Any Variable/Attribute with RegEx
• Function Entry
• Variables in Specified Function
• Garbage Collector Operation
• Function Input Arguments
• Function Return Value
• Audit Events
• Raised Exceptions

Add Custom Event

VizTracer supports inserting custom events while the program is running. This works like a print debug, but you can know when this print happens while looking at trace data.

• Instant Event
• Variable Event
• Duration Event

Misc

Multi Thread Support

VizTracer supports python native threading module without the need to do any modification to your code. Just start VizTracer before you create threads and it will just work.

For other multi-thread scenarios, you can use enable_thread_tracing() to notice VizTracer about the thread to trace it.

Refer to multi thread docs for details

Multi Process Support

VizTracer supports subprocess, multiprocessing, os.fork(), concurrent.futures, and loky out of the box.

For more general multi-process cases, VizTracer can support with some extra steps.

Refer to multi process docs for details

Async Support

VizTracer supports asyncio natively, but could enhance the report by using --log_async.

Refer to async docs for details

Flamegraph

Perfetto supports native flamegraph, just select slices on the UI and choose "Slice Flamegraph".

Remote attach

VizTracer supports remote attach to an arbitrary Python process to trace it, as long as viztracer is importable

Refer to remote attach docs

JSON alternative

VizTracer needs to dump the internal data to json format. It is recommended for the users to install orjson, which is much faster than the builtin json library. VizTracer will try to import orjson and fall back to the builtin json library if orjson does not exist.

Performance

VizTracer will introduce 2x to 3x overhead in the worst case. The overhead is much better if there are less function calls or if filters are applied correctly.

An example run for test_performance with Python 3.8 / Ubuntu 18.04.4 on Github VM

fib:
0.000678067(1.00)[origin]
0.019880272(29.32)[py] 0.011103901(16.38)[parse] 0.021165599(31.21)[json]
0.001344933(1.98)[c] 0.008181911(12.07)[parse] 0.015789866(23.29)[json]
0.001472846(2.17)[cProfile]

hanoi     (6148, 4100):
0.000550255(1.00)[origin]
0.016343521(29.70)[py] 0.007299123(13.26)[parse] 0.016779364(30.49)[json]
0.001062505(1.93)[c] 0.006416136(11.66)[parse] 0.011463236(20.83)[json]
0.001144914(2.08)[cProfile]

qsort     (8289, 5377):
0.002817679(1.00)[origin]
0.052747431(18.72)[py] 0.011339725(4.02)[parse] 0.023644345(8.39)[json]
0.004767673(1.69)[c] 0.008735166(3.10)[parse] 0.017173703(6.09)[json]
0.007248019(2.57)[cProfile]

slow_fib  (1135, 758):
0.028759652(1.00)[origin]
0.033994071(1.18)[py] 0.001630461(0.06)[parse] 0.003386635(0.12)[json]
0.029481623(1.03)[c] 0.001152415(0.04)[parse] 0.002191417(0.08)[json]
0.028289305(0.98)[cProfile]

Documentation

For full documentation, please see https://viztracer.readthedocs.io/en/stable

Bugs/Requests

Please send bug reports and feature requests through github issue tracker. VizTracer is currently under development now and it's open to any constructive suggestions.

License

Copyright 2020-2024 Tian Gao.

Distributed under the terms of the Apache 2.0 license.